Built for the CLI
Create projects, version prompts and deploy through the CLI, with structured JSON output your agent can work with.
Prompt management,
on autopilot.
Monitor performance, evaluate responses, and update prompts & models automatically.
to start free. No credit card.
Setup
One prompt. You're connected.
and paste this into your agent.
Use the PromptOn setup skill: https://app.prompton.ai/docs/skills/prompton-setup/SKILL.md Explain what PromptOn is and how it works. Check whether it's a good fit for this codebase. If it is, propose a plan to migrate this project to PromptOn. Proceed with the migration only after I explicitly approve the plan.
What happens during setup?
-
01
Explain & check fit
The agent tells you what PromptOn is, makes sure it can see your code, and finds every LLM call. If PromptOn isn't right for your app, it says so and stops.
-
02
Install & login
The CLI installs in one line. We email you a six-digit code; you sign in, then approve the login. The agent waits while you do.
-
03
Compare & plan
The AI compares how your app manages prompts today with the proposed setup, then provisions matching prompts in your lowest active nonproduction environment before any app code migration.
-
04
Set up safely
The AI records PTN_SETUP_ENV, creates prompts, deploys them there and sends ten sample monitoring logs per prompt first. It shows the prompts and Logs pages, then asks before migrating app code in a new git worktree. It shows the code diff, monitoring path and verification results, then stops before production.
Your coding AI follows the setup skill, falls back to the agent guide, and verifies PTN_SETUP_ENV before any production rollout.
Skip the clicks. Let your AI handle it.
Agent-friendly by design.
One page it can read
The setup skill covers the onboarding workflow, with the full agent guide available when more detail is needed.
Read the setup skillYou stay in control
Your agent checks the fit, shows the migration plan and waits for your approval before applying changes.
Why PromptOn
Prompt quality won’t
maintain itself.
Prompts in code or a database can get your app running. Keeping them at their best is the work that never stops.
The problem
Prompts in your code
Every improvement
waits for a release.
A new prompt or model means a code change, a review and another app release—before you even know if it works better.
Prompts in your database
You built a prompt table.
Now build the tooling.
Editing gets easier. But version history, monitoring and evaluations are still tools your team has to build and maintain.
Either way, your team owns the loop.
The solution · PromptOn
New prompt. New model.
Same app code.
PromptOn gives your app the prompt, model and parameters for each LLM call. Compatible updates happen here, without another app release.
- Version history
- Response monitoring
- Evaluations
# New model or prompt?
# Edit this code and redeploy the app.
answer = client.chat.completions.create(
model="gpt-4o",
temperature=0.3,
messages=[
{"role": "system", "content": instructions},
{"role": "user", "content": question},
],
)# No code edits. No app redeploy.
# Change the model or prompt at https://app.prompton.ai
prompt = prompton.prompt("support_reply")
answer = client.chat.completions.create(
model=prompt.model,
**prompt.params,
messages=prompt.messages({"question": question}),
)Simplified example. Your app calls the provider directly. Full integration →
Features
From evaluation to updates. Automatically.
Evaluate the responses that matter.
Evaluate real responses against your prompt’s criteria, from policy accuracy to tone and useful next steps.
Prompt template
System
You are a support agent for {{account_name}}.
Tone: {{tone}}
Policy: include refund window and next step.
User
{{customer_question}}
Evaluation queue
Refund question
Billing clarification
Update prompts and models.
Update the prompt, model and parameters for the same prompt. Compatible changes leave your app code untouched.
Before · v17
- Model
- Model A
- Prompt
- friendly support agent
- Params
- temperature 0.3
After · v18
- Model
- Model B
- Prompt
- policy-aware support agent
- Params
- temperature 0.2
Every change, recorded.
Trace each response to the prompt and model behind it. Compare versions or roll back when you need to.
v18
Policy and model updated
setup env
v17
Earlier prompt and model
previousSee what happens next.
Track response quality, latency and cost after each change, tied to the version that produced the response.
Refund answer
v18
820 ms
$0.004
Billing answer
v18
760 ms
$0.003
In your app
We're not in the request path.
- PromptOn supplies the prompt, model, and parameters your app uses to make LLM calls.
- Your app calls the model provider directly with your own API keys.
Your app sends requests directly to the model provider with your API keys and receives its responses. PromptOn supplies prompt and model configuration to your app and receives batched logs for monitoring and automated evaluation. There is no connection between PromptOn and the model provider in this request path.
Pricing
Start free. Grow when you're ready.
Plans for personal work and growing teams.
-
Free Available now
$0
Get started today. No credit card.
-
Team
$29/ month
More monitoring capacity and shared access.
-
Pro
$99/ month
Automatic evaluation and longer metadata history.
FAQ
A few more things.
Can I self-host PromptOn?
PromptOn is available as a hosted service. The SDKs and CLI are open source under Apache 2.0. View the SDKs and CLI on GitHub →
Is this a proxy?
No. Your app calls the model provider directly with your own keys. PromptOn only serves configuration and receives logs.
Which providers?
OpenRouter today, which covers most models. More are coming.
Where do my prompts live?
In PromptOn, on an encrypted database. Monitoring-log content and provider keys are additionally encrypted with a separate key. Export anytime.